Re: [PATCH v50 0/11] MR10567: wined3d: fix SRGB RTV fail for UNORM swapchain backbuffers > 1 for d3d10/11

[Stian Low (＠stianlow)]

> This is hard to read and could probably do with quite a lot of splitting
> Tracking the list of swapchain views, while not doing anything with them yet, can be its own commit.
> GL and Vulkan rotate implementations should ideally be separated.

Commits have been split.  Please let me know if they should be adjusted more.
> D3D11_FORMAT_SUPPORT_BACK_BUFFER_CAST implementation can be its own commit It also should not use a fixed list of formats, and should live in wined3d like all the other format flags.

D3D11_FORMAT_SUPPORT_BACK_BUFFER_CAST now handled along with the other format flags for `ID3D11Device_CheckFormatSupport
> D3D11_FORMAT_SUPPORT_BACK_BUFFER_CAST validation similarly can be its own commit, and also should live in wined3d.  Actually, you seem to unnecessarily have it in both places?

`wined3d_check_device_format` is called directly now instead of via `ID3D11Device_CheckFormatSupport` for less `d3d_rendertarget_view_init` overhead which handles the validation specifically for d3d10/11 all in one call.

`ID3D11Device_CheckFormatSupport` changes are technically no longer needed to support the fix but included with patchset anyway since it may be needed to fix other apps.
> I would not use the fixed array size of 29; just allocate it dynamically instead the same as we do for the actual list of backbuffers. With that done, wined3d_swapchain_rendertarget_view_gl is fairly unnecessary; it's just a single pointer that can be put directly in wined3d_rendertarget_view like the others.

`wined3d_swapchain_rendertarget_view_gl` and vk have been changed as list structs instead of wrappers of `wined3d_rendertarget_view_gl` and initialized based on `view_gl->bo_user.entry`.
> Disabling GL_FRAMEBUFFER_SRGB on swapchain blit is still wrong, as Henri said. (Moving it back "to fix test failures" is not correct; you need to understand why it should be there.) And that should be its own commit; it's not related to the rest of this.

GL_FRAMEBUFFER_SRGB has been moved back to `texture2d_blt_fbo` with a condition for `resolve_format` `WINED3D_FORMAT_CAP_SRGB_WRITE`.  Unclear if this is actually correct but it fixes tests and Frostpunk.  @zfigura originally mentioned WINED3D_FORMAT_CAP_SRGB_WRITE may be a discrepancy before I discovered lack of backbuffers support was a wider problem so it seems to come full circle.

Some extra vulkan/damavand tests fail locally despite wine server tests passing.  Is there a way to get wine server to also test vulkan?  Vulkan may have bugs that this patch exposes similar to GL_FRAMEBUFFER_SRGB bug.

I think this merge is ready for another review while vulkan fails are addressed.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/10567#note_144391